Introduction: Accurate forecasting of rainfall is essential in modern flood prediction systems. Among various meteorological models, the Weather Research and Forecasting (WRF) model is recognized as one of the most reliable mesoscale models for simulating precipitation in diverse climatic and geographic conditions. However, like many numerical models, WRF is subject to temporal and spatial uncertainties, particularly in regions with complex topography such as western and southwestern Iran. This study evaluates the accuracy of WRF’s rainfall predictions across multiple storm events that resulted in significant flooding in these regions between 2014 and 2016.
The primary objective of this research is to assess the spatial and temporal accuracy of rainfall forecasts produced by the WRF model over lead times of 24, 48, 72, and 96 hours. The study specifically aims to quantify overestimation or underestimation errors and determine how these errors evolve with increased forecast horizon.
Methodology: Five major storm events associated with significant flooding were selected for evaluation. These storms occurred in the western and southwestern basins of Iran, which are prone to intense rainfall and flash flooding due to their topographic and climatic features. WRF model outputs for precipitation were obtained for each event at different lead times (24–96 hours). Observational rainfall data from 100 synoptic stations operated by the Iranian Meteorological Organization were used as ground truth.
The spatial distribution of observed and predicted rainfall was analyzed using Geographic Information System (GIS) tools, and statistical indices such as Mean Error (ME), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Bias were used to validate the model’s performance. Time series comparisons and hydrographs were also produced to compare observed and modeled rainfall for selected stations within the basins.
Results and Discussion: The results indicate that the WRF model tends to overestimate rainfall in the majority of stations, while in a smaller number of cases it underestimates it. The extent of overestimation was more significant in short-term forecasts (24 hours), with an average over-prediction error of around 30%. Interestingly, this error gradually decreased with increasing forecast horizon, reaching about 25% at the 96-hour lead time.
Spatially, the highest discrepancies were observed in stations located in mountainous areas, where orographic effects and convective activity are more pronounced and harder to simulate. The hydrograph comparison for two major dams—Karun 4 and Dez—demonstrated that although the model captures the timing and general shape of the flood hydrograph, it often overestimates the peak discharge, especially in the shorter forecast windows.
These findings emphasize the importance of calibrating model outputs and incorporating ensemble forecasting or data assimilation techniques to improve the reliability of flood warnings based on WRF outputs.
Conclusion: This study confirms that while the WRF model provides valuable forecasts for flood prediction applications, there is a consistent pattern of rainfall overestimation, particularly at shorter lead times. However, the decline in error with extended lead times suggests model stability in longer forecasts. These insights can be used to enhance early warning systems in Iran by adjusting model outputs or integrating them with local observations and hydrological models such as HEC-1 for flood routing.